Create model

Description

The model for localization will be created with the registered iBeacons and the fingerprinting data. Then the model will be trained by a machine learning tool. This file will be downloaded by NavCog app.

Pre-requisite

• LocationService server (see here)

Steps

1. compose model file for NavCog3
2. training the model file
3. upload the model to MapService server

1. Compose model file for NavCog3

• open http://<location-service-host>/LocationService/data_composer.html
• select floorplans you want to export
• select an anchor for the export
• click "Export"

About map matching images

If you want to get more accurate localization, please consider to provide images for map matching.

• You can get a simple map matching images based on your path network by specifying path network geojson file for the data composer (see prepare data for routing)
• The format of the map matching image is described here
• You can edit the composed data at http://<location-service-host>/LocationService/2dMapData/

2. Training the model

• train model parameter by BasicLocalizer, a command line tool of blelocpp

$ BasicLocalizer -m model.json --train

Tips for training model

• this step is required for localization
• please check the values of stdev in the output of the tool. Those are usually in between about 2 to 4.

start setModel
parse JSON: 4ms
loading
...
stdev(2017,1) = 2.20419
stdev(2017,2) = 2.58539
stdev(2017,3) = 2.25654
stdev(2017,4) = 2.70081
stdev(2017,5) = 2.78212
stdev(2017,6) = 3.00939
...

3. Upload the model

• upload model file (see here)
• specify model file in server config file (see here)